Inside Fable 5: the features a solo practitioner should actually turn on.

Most practitioners who try Fable 5 use one percent of it. They open the chat box, ask a clinical question, get a competent paragraph, and conclude it is a fancier textbook. That is a fair verdict on the box. It is a completely wrong verdict on the tool, because the box is the part that was already there. The associate you could never afford is built out of the features underneath it — and most of them are switched off by default because nobody told you to turn them on.

So here is the induction, feature by feature. Not a settings walkthrough — the small set of capabilities that actually change what a one-person practice can carry, and the exact point in each where a licensed human has to stay in the loop.

feature one: the context window means it can hold the whole client

The large context window is not a spec to admire — it is the reason this behaves like an associate instead of a temp. A small model forces you to re-summarise the client for every task, which is so tiring that people give up and use it for generic reading. A large-context model holds the intake form, the last three sessions, the labs and the goals in one pass, and reasons across all of them together. The session prep knows what the intake said. The follow-up knows what you actually agreed.

The move is to brief once, not repeatedly. Write the model a proper picture of the client at the start — the way you would hand an associate the file before a handover — and then delegate task after task against it. You are not writing cleverer prompts. You are refusing to re-introduce the client every single time.

feature two: uploads — let it read the file, not your summary

Uploads are where the quality jumps. Fable 5 can read the intake PDF, the lab panel, the food diary, the client’s own screenshotted wearable data. Instead of you typing a lossy summary and asking it to reason about that, you hand it the source and ask it to do the reading. The intake tournament gets sharper when the model is working from the raw form rather than your paraphrase of it.

Give it the raw intake and ask for three different framings of the presenting picture — different emphases, different hypotheses about what is driving what — then have it argue for the most clinically useful one and flag what each might be missing. You are not outsourcing the diagnosis. You are generating more angles than a tired brain reaches at 9pm, and then choosing between them with your own judgement.

feature three: projects — a room per client, briefed once

A project is a persistent workspace with its own instructions and files. For a practice this is the feature that quietly changes everything: a room per client, or a room per workflow. Set up one project with your standing brief — how you practise, what a good summary looks like to you, your tone, your non-negotiables — and every task you run inside it inherits that context without you retyping it.

The highest-leverage version is to have the project interview the file before a session. ‘What are the ten questions this history raises that I have not resolved? What contradicts what? What did I say I would follow up on and never did?’ It is not telling you what to do. It is doing what a good colleague does over coffee — asking the questions that make you sharper before you walk into the room. That is how you protect attentiveness as your caseload grows: care degrades at volume because of memory, not effort, and a project that re-reads the whole file before every session gives you back the memory.

feature four: scheduled actions and connectors — the quiet automation

Two features finish the associate. Scheduled actions let you set recurring work — a weekly sweep of who is due a follow-up, a draft check-in for clients you have not seen in a month. Connectors let the model reach into the tools you already run — your calendar, your inbox, your forms — so it drafts against live information instead of a snapshot you pasted in. Turned on carefully, they close the gap between ‘I should follow up with them’ and it actually happening.

Turn these on last, and turn them on narrow. A connector that can read is far safer than one that can send. Let the model draft the follow-up and surface who needs one; keep the sending, and the clinical call inside it, with you. The automation should remove the remembering, never the deciding.

“The associate you could never afford is not the chat box. It is the memory, the uploads, the room, and the clock — assembled around the one thing that stays a human: your sign-off.”

the client-facing rule that keeps you safe

Two commitments hold all of this together as you turn features on. First, the model drafts and you decide — nothing reaches a client without a human clinician’s sign-off, however good the draft. Second, be honest with clients about where AI helps in your process and where your judgement is doing the work; that transparency is becoming a trust advantage, not a liability. Your value was never that you could remember everything. It was that you could be trusted to decide. These features make the remembering cheap so you can spend more of yourself on the deciding.

what to turn on this week

Do not turn everything on at once. Pick one project and one feature. Create a room for your next new client, upload the raw intake, and delegate three things against your standing brief: the summary, the session prep, and a first-draft follow-up. Read each as a supervisor reading a junior’s work — correct it, keep the good, bin the rest. Only once that feels reliable should you add a scheduled sweep or a read-only connector.